Research in Safe Autonomy & Robot Safety at the University of Michigan ensures that robots, whether working beside humans every day or exploring Mars, operate correctly and safely. Because testing every possible situation a robot might face is impossible, faculty use verification and validation (V&V) to provide mathematical certificates that a system will follow its safety rules. This work involves creating specification languages and modeling formalisms to describe correct behavior in a way that computer programs can analyze for “corner cases” or errors before they happen.
A major focus is correct-by-construction control synthesis, which is essentially using “software to design other software” so that the resulting robot is guaranteed to follow its instructions. To keep robots safe in the real world, researchers use reachability-based trajectory design and control barrier functions. These methods create a zone of safety around the robot that can automatically override a command to prevent a crash or a fall.
For systems to be truly reliable, they must have fault-tolerance and monitoring to check their own health at run-time. Michigan engineers build prognostics systems and digital twins that run simulations in parallel with the real robot to catch anomalies or detect when sensors have been compromised. This includes adversarial resilience, which helps robot teams stay safe even if they are attacked or sent bad data. Additionally, researchers address uncertainty by teaching robots to understand when their own models are unreliable, ensuring they can handle messy or unpredictable environments without causing accidents.
